rag-memory-epf-mcp 3.5.0 → 3.5.2

package/dist/index.d.ts

@@ -111,6 +111,7 @@ export declare class RAGKnowledgeGraphManager {
     searchNodes(query: string, limit?: number, since?: string, until?: string): Promise<KnowledgeGraph>;
     openNodes(names: string[]): Promise<KnowledgeGraph>;
     private generateEntityEmbeddingText;
+    private buildEntityEmbeddingText;
     private splitIntoSentences;
     private calculateSentenceSimilarities;
     private cosineSimilarity;

package/dist/index.js

@@ -638,12 +638,49 @@ export class RAGKnowledgeGraphManager {
         return { entities, relations };
     }
     // === NEW RAG FUNCTIONALITY ===
-    // Generate embedding text for an entity (combines name, type, and observations)
+    // Generate embedding text for an entity (identity + newest observations within a char budget).
     generateEntityEmbeddingText(entity) {
-        const observationsText = entity.observations
-            .filter(o => !o.startsWith('Source:') && !o.startsWith('Created:') && !o.startsWith('Type:') && !o.startsWith('Tags:') && !o.startsWith('Content length:'))
-            .join('. ');
-        return `${entity.entityType}: ${entity.name}. ${observationsText}`.trim();
+        return this.buildEntityEmbeddingText(entity).text;
+    }
+    // Build entity embedding text plus stats for instrumentation.
+    // The char budget keeps the entity vector representative of CURRENT state and stays under the
+    // bge-m3 8192-token ceiling; older history lives in RAG document chunks / dated entities.
+    // Returns selected/total observation counts and filtered (pre-cap) vs capped observation char
+    // sizes so callers can log truncation accurately (identity prefix is excluded from these sizes).
+    buildEntityEmbeddingText(entity) {
+        const maxObservationChars = Math.max(1000, Number.parseInt(process.env.ENTITY_EMBED_OBS_CHAR_BUDGET || '12000', 10) || 12000);
+        const observations = entity.observations.filter(o => !o.startsWith('Source:') && !o.startsWith('Created:') && !o.startsWith('Type:') &&
+            !o.startsWith('Tags:') && !o.startsWith('Content length:'));
+        const filteredObsChars = observations.join('. ').length;
+        const selected = [];
+        let remaining = maxObservationChars;
+        for (let i = observations.length - 1; i >= 0 && remaining > 0; i--) {
+            const obs = observations[i];
+            const separatorCost = selected.length > 0 ? 2 : 0; // '. ' joiner
+            const available = remaining - separatorCost;
+            if (available <= 0)
+                break;
+            if (obs.length <= available) {
+                selected.push(obs);
+                remaining -= obs.length + separatorCost;
+            }
+            else if (selected.length === 0) {
+                selected.push(obs.slice(0, available)); // single giant obs: keep a truncated head, never empty
+                break;
+            }
+            else {
+                break;
+            }
+        }
+        const observationsText = selected.reverse().join('. ');
+        const text = `${entity.entityType}: ${entity.name}. ${observationsText}`.trim();
+        return {
+            text,
+            filteredObsChars,
+            cappedObsChars: observationsText.length,
+            selectedObsCount: selected.length,
+            totalObsCount: observations.length,
+        };
     }
     // NEW: Generic semantic summary generation methods
     splitIntoSentences(text) {
@@ -775,13 +812,19 @@ export class RAGKnowledgeGraphManager {
             return false;
         }
         const parsedObservations = JSON.parse(entity.observations);
-        const embeddingText = this.generateEntityEmbeddingText({
+        const built = this.buildEntityEmbeddingText({
             name: entity.name,
             entityType: entity.entityType,
             observations: parsedObservations
         });
-        // Generate embedding
+        const embeddingText = built.text;
+        // Instrumentation (stderr only): observations kept vs total, filtered pre-cap vs capped obs
+        // char size (identity excluded), and embed duration. `capped` = some observation chars dropped.
+        const capped = built.cappedObsChars < built.filteredObsChars;
+        const embedStart = Date.now();
         const embedding = await this.generateEmbedding(embeddingText);
+        const embedMs = Date.now() - embedStart;
+        console.error(`[embed] ${entity.name}: ${built.selectedObsCount}/${built.totalObsCount} obs, ${built.filteredObsChars}ch -> ${built.cappedObsChars}ch${capped ? ' (capped)' : ''}, ${embedMs}ms`);
         try {
             // Delete existing embedding if any
             const existingMetadata = this.db.prepare(`
@@ -2660,9 +2703,12 @@ async function main() {
         process.exit(1);
     }
 }
-// Only boot the server when run as the entry point — not when imported (tests).
-const isDirectRun = process.argv[1] === fileURLToPath(import.meta.url);
-if (isDirectRun) {
+// Boot the server unless explicitly suppressed. Tests import this module with
+// RAG_MEMORY_NO_AUTOSTART=1 to access the class without starting the stdio server.
+// (An argv-vs-import.meta.url comparison is unreliable: npx/bin launches the entry
+// via a symlinked path, so the two never match and main() silently skips → the
+// MCP client cannot connect. v3.5.0 shipped that bug; env-var opt-out is robust.)
+if (process.env.RAG_MEMORY_NO_AUTOSTART !== '1') {
     main().catch((error) => {
         console.error("Fatal error in main():", error);
         ragKgManager.cleanup();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rag-memory-epf-mcp",
-  "version": "3.5.0",
+  "version": "3.5.2",
   "description": "Project-local RAG memory MCP server — knowledge graph + multilingual vector + FTS5 in a single SQLite file. Per-project isolation, 30 MCP tools, codepoint-safe chunking (Korean/CJK/emoji).",
   "keywords": [
     "mcp",
@@ -41,7 +41,7 @@
     "prepare": "npm run build",
     "watch": "tsc --watch",
     "verify:invariants": "node test/chunk-invariants.test.mjs",
-    "verify:engine": "node test/engine-smoke.test.mjs && node test/sync-atomicity.test.mjs && node test/dedup.test.mjs && node test/search-degradation.test.mjs",
+    "verify:engine": "node test/engine-smoke.test.mjs && node test/launch-smoke.test.mjs && node test/sync-atomicity.test.mjs && node test/dedup.test.mjs && node test/search-degradation.test.mjs && node test/entity-embed-cap.test.mjs",
     "test": "npm run build && npm run verify:invariants && npm run verify:engine",
     "prepublishOnly": "npm run build && npm run verify:invariants && npm run verify:engine"
   },